WAV to text
WAV is uncompressed, so it is the cleanest thing you can hand a speech model — nothing has been thrown away by an encoder. Drop one in and it is transcribed on your device.
It is also the format that fixes almost every "this file will not decode" problem: exporting an awkward recording to WAV in any free editor makes it readable everywhere.
Language and translation need the Multilingual model.
Drop an audio or video file
or , paste with Ctrl + V, or
- MP3
- MP4
- WAV
- M4A
- MOV
- WEBM
- OGG
- FLAC
- AAC
The file is read by this page and never uploaded — no account, no size limit but your own RAM.
Preset: balanced English model, plain-text output.
Download
Click a line to jump the audio there. Click the text to fix a word — your edits go into every download.
How it works
- Open this page — the tool is already set up for “WAV to text”.
- Drop the file in, browse for it, or paste it with Ctrl+V.
- The model runs on your own device; the transcript appears as it goes.
- Correct anything misheard, then download text, subtitles or notes.
WAV as the universal fallback
Browsers do not ship decoders for everything. WMA, AMR, some MKV variants and a handful of proprietary dictation formats simply have no decoder available in a browser tab, and no amount of client-side cleverness changes that. Converting to WAV with Audacity, ffmpeg or any free converter turns an unreadable file into one that works everywhere — including here.
If you are converting anyway, 16-bit mono at 16 kHz is all this needs and produces the smallest file. Higher rates are fine and are downsampled automatically; they just make the file larger for no gain in accuracy.
The size trade-off
Uncompressed audio is big: roughly 10 MB per minute at CD quality stereo, or about 2 MB per minute at 16 kHz mono. That affects how long the file takes to read off disk, not how long transcription takes — the model works from decoded samples either way, so a WAV and an MP3 of the same recording run at the same speed.
For very long WAV recordings the memory footprint is what to watch. If a multi-hour session makes the tab struggle, converting to 16 kHz mono first cuts the memory needed by roughly six times without touching accuracy.
Multi-track and interview recordings
Field recorders often produce a stereo or multi-channel WAV with a different microphone per channel. Channels are averaged into mono here, so every voice is present — you do not lose the person on channel two. That is deliberate: taking only the first channel is a common shortcut that silently drops half of an interview.
If you have genuinely separate per-speaker files, transcribing each one on its own and interleaving by timestamp gives you real speaker separation, which is more reliable than any automatic diarization.
Things that save a re-run
- Converting to 16 kHz mono WAV before a very long run saves a lot of memory and costs no accuracy.
- If a file will not decode anywhere, WAV is the format to convert to — it is the lowest common denominator.
- Normalise quiet recordings before transcribing; the model does not apply automatic gain.
Read the transcript against the audio before you rely on it. Every speech model — this one and the paid cloud ones — mishears names, numbers and crosstalk, and it does so confidently. The player and editor here exist so that check takes minutes rather than an afternoon.
Frequently asked questions
Is WAV more accurate than MP3?
Marginally, and usually not enough to notice. The model works from a 16 kHz mono spectrogram, which a decent MP3 already preserves. WAV matters when the original is WAV — re-encoding an MP3 to WAV adds nothing back.
What sample rate should I use?
16 kHz mono is exactly what the model consumes, so it is the efficient choice. Anything higher is downsampled automatically and works fine; anything lower loses information the model would have used.
My WAV file is several gigabytes — will it work?
It depends on your memory, because the decoded audio is held in RAM. Converting to 16 kHz mono first reduces that dramatically. If the tab still struggles, split the file and transcribe the parts.
Are multi-channel recordings handled?
Yes. All channels are averaged into mono, so every microphone contributes. Nothing is dropped, which is what would happen if only the first channel were used.